Inflation without a beginning: a null boundary proposal
Anthony Aguirre
School of Natural Sciences, Institute for Advanced Study Princeton, New Jersey 08540, USA

Steven Gratton
Joseph Henry Laboratories, Princeton University, Princeton, New Jersey 08544, USA

(Dated: February 7, 2008)
We develop our recent suggestion that inflation may be made past eternal, so that there is no initial
cosmological singularity or “beginning of time”. Inflation with multiple vacua generically approaches
a steady-state statistical distribution of regions at these vacua, and our model follows directly from
making this distribution hold at all times. We find that this corresponds (at the semi-classical level)
to particularly simple cosmological boundary conditions on an infinite null surface near which the
spacetime looks de Sitter. The model admits an interesting arrow of time that is well-defined and
consistent for all physical observers that can communicate, even while the statistical description
of the entire universe admits a symmetry that includes time-reversal. Our model suggests, but
does not require, the identification of antipodal points on the manifold. The resulting “elliptic” de
Sitter spacetime has interesting classical and quantum properties. The proposal may be generalized
to other inflationary potentials, or to boundary conditions that give semi-eternal but non-singular
PACS numbers: 98.80.Cq, 04.20.Gz, 04.62.+v
The ancient philosophical question of whether the uni-
verse is finite or infinite in time, and whether time “had
a beginning”, entered the domain of scientific study dur-
ing the 20th century with the development of generally
relativistic cosmologies in which the universe is homoge-
neous and expanding, as observed on the largest accessi-
ble scales.
Chief among these cosmologies, and representing oppo-
site answers to the question of the universe’s beginning,
were the classical Big-Bang and Steady-State. As FRW
models, both are based on some form of the “Cosmolog-
ical Principle” (CP) that the large-scale statistical prop-
erties of the universe admit spatial translational and ro-
tational symmetries. The models differ greatly, however,
in their time evolution. In the Big-Bang, the properties
of the universe evolve in a finite time from a dense, singu-
lar initial state. In contrast, the Steady-State universe is
said to obey the “Perfect Cosmological Principle” (PCP)
in that it admits, in additional to spatial translational
and rotational symmetries, a time-translation symmetry.
Since all times are equivalent, there can be no “beginning
of time”, and the universe is infinite in duration.
Unlike their philosophical predecessors, the Big-Bang
and Steady-State model were observationally distinguish-
able, and astronomical evidence eventually turned nearly
all cosmologists away from the Steady-State. Moreover,
theorems proven within General Relativity showed that
the classical singularity of the Big-Bang cosmology was

Electronic address: aguirre@ias.edu

Electronic address: sgratton@princeton.edu
robust and could not be avoided by relaxing simplify-
ing assumptions such as that of homogeneity. Thus the
idea of a temporally finite universe with a singular initial
epoch came to dominate cosmology.
Attention has since
focused on how this primordial singularity (where some
presently unknown theory of quantum gravity presum-
ably applies) could give rise to a classical “initial” state
that could evolve into the observed universe.
The required “initial” classical state, however, seemed
rather special: the universe had to have been extremely
flat, and statistically homogeneous (i.e. obey the CP)
on scales larger than the horizon size. The theory of
inflation was devised and widely accepted as a solution
to this problem of a special initial state: given inflation,
a flat, homogeneous universe (with the necessary Gaus-
sian scale-invariant density fluctuations) is an attractor.
That is, within some inflating region of fixed, finite phys-
ical size, the CP holds more and more precisely with
time. What is perhaps more surprising and less widely
appreciated, however, is that in generic inflation models
the universe also comes to obey, with ever-greater preci-
sion, the Perfect Cosmological Principle. This occurs be-
cause inflation is generically “semi-eternal”: rather than
ending globally at some time, inflation always continues
in some regions, and the universe globally approaches a
quasi-steady-state distribution of inflating and thermal-
ized regions, the statistical description of which becomes
There have been a number of proposals for avoiding a beginning
of time, but generally these involve either continuing through
the cosmological singularity by invoking quantum gravitational
effects, or modifying GR at the classical level. Our approach aims
to develop a non-singular cosmology without appeal to either
asymptotically independent of time [1].
Since inflation generically approaches a steady-state, it
seems physically reasonable to ask whether the universe
can simply be in an inflationary steady-state, thus avoid-
ing a cosmological singularity or “beginning of time”. In-
deed, the possibility of truly eternal inflation was raised
soon after inflation’s invention, but no satisfactory model
was immediately devised [2], and in subsequent years
several theorems were formulated proving the geodesic
incompleteness of models globally satisfying conditions
seeming necessary for eternal inflation [3, 4]. These the-
orems suggested that inflationary cosmologies necessarily
contain singularities, but the exact nature of the implied
singularities was obscure.
In a recent paper, we constructed a counter-example
to these theorems by providing a model for geodesically
complete truly eternal inflation [5]. There, we analyzed
the classical Steady-State model in detail, then extended
our analysis to inflation. Here, we develop the model
from a different standpoint, focusing on an eternal infla-
tion in a double-well inflaton potential, and on the cor-
responding cosmological boundary conditions. Section II
motivates and develops our model, and describes its gen-
eral features. Various aspects of the model are developed
in subsequent sections: Section III discusses the arrow of
time in our model, and elucidates the failure of the sin-
gularity theorems to forbid our construction; Sec. IV dis-
cusses the cosmological boundary conditions, which are
specified on a null surface; Sec. V discusses the relation
of our model to the “antipodally identified” or “elliptic”
interpretation of de Sitter spacetime that it suggests, and
briefly discusses quantum field theory in elliptic de Sit-
ter; Sec. VI discusses generalizations and extensions of
our model. We summarize and conclude in Sec. VII.
In this section we develop an eternally inflating cos-
mology based on a double-well inflaton potential U(φ)
with minima at φ
and φ
, where U(φ
) > U(φ
) ≥ 0.
This sort of potential is posited in “old” inflation [6] or
“open” inflation [7, 8]. We will first review semi-eternal
double-well inflation, then extend this to eternal infla-
tion, then analyze and address the geodesic completeness
of the model.
A. Semi-eternal “double-well” inflation
A semi-eternally inflating cosmology naturally arises
from a generic double-well potential [9, 10]. Consider
some large comoving region in which, at an initial time
t = t
, the energy density is dominated by the inflaton
φ, with φ ≃ φ
and ∂
φ ≃ 0. Assume the spacetime to
locally resemble de Sitter (“dS”) space, with (for conve-
nience) nearly-flat spatial sections, i.e. with a metric [11]
approximated by
= −dt


. (1)
In this background, bubbles of true vacuumφ
at a fixed rate λ per unit physical 4-volume that depends
upon the potential U(φ)[12]. The interior of each bubble
looks like an open FRW cosmology to observers inside it.
For a suitably designed U(φ) (as in open inflation[7, 8]),
there can be a slow-roll inflation epoch inside the bub-
ble so that the FRW regions are nearly flat and homo-
geneous, and have scale-invariant density perturbations.
One such region could therefore in principle represent our
observable cosmological surroundings.
At any time t, we can derive the distribution of bub-
bles and inflating region within our comoving volume,
with the aim of showing that the distribution approaches
a steady-state. To avoid complications resulting from
bubble collisions and the ambiguities in connecting the
time-slicings within and outside bubbles, we concentrate
on the statistics describing the inflating region outside
of the bubbles. This region is necessarily unaffected by
the bubbles’ presence because they expand at the speed
of light: both its global and local properties depend only
on its initial state at t
. But we may describe three effects
of the bubble encroachment upon it.
First, let us consider the inflating region left at time
t by bubbles forming since t
. This region must consist
entirely of points each of which does not have a nucle-
ation event in its past light-cone (PLC) going back to t
Denoting the volume of this PLC by Q(t, t
), it can then
be shown that such points comprise a volume fraction
= exp[−λQ] ≃ exp
−4πλ(t −t

for (t − t
) ≫ H
[9, 10]. Although f
→ 0 for large
t − t
, the spacetime is said to be eternally inflating be-
cause for small λ the physical inflating volume within our
comoving region nonetheless increases exponentially with
∝ f
exp(3Ht) ∼ exp(DHt) (3)
for any fixed t
, with D ≡ 3 −4πλ/3H
Second, one can show [9] that at fixed t the distribution
of inflating regions about any inflating point is described
by a fractal of dimension D (that is, the inflating volume
∝ r
(r) ∝ r
) up to a scale of order r
(t, t
(t, t
) = H


is the physical radius at t of a bubble nucleated at t
Third, we may calculate, for a given point in the in-
flating region at time, the number per unit time N(r, t)
of incoming bubbles of physical radius r. This is
N(r, t) =
(1 +Hr)
FIG. 1: Conformal diagram for de Sitter (dS) spacetime.
Each point represents one point in 1+1 D dS, or half of a
2-sphere in 3+1 D dS. The left and right (dotted) edges are
identified. The shaded region (region I) is covered by coor-
dinates with flat spatial sections (spacelike lines) with space-
like infinity at i0; the straight, timelike lines represent co-
moving geodesics. The null surface J

represents t → −∞.
True-vacuum “test bubbles” (not disturbing the background
spacetime) are darkly shaded and open toward future time-
like infinity J
. Also shown are are the light-cones of points
P and Q in regions I and II that open toward J

, and null
(“Y”) and timelike (“X”) geodesic segments with an endpoint
at xi and crossing J

for r < r
(t, t
) and zero for r > r
An observer within a bubble can never leave, but will
eventually be encountered by an encroaching bubble wall
after a typical time τ
, where τ
is related to the r-
integral of Eq. (5) by some transformation between the
bubble observer’s proper time τ and cosmic time t. Since
this rate depends on t − t
, a patient and very sturdy
observer could in principle discover the global time at
which it formed by counting the frequency of incoming
Now, as t → ∞, four things occur. First, the nearly-
flat spatial sections approach perfect flatness. Second,
the incoming bubble rate N(r, t) becomes homogeneous
and independent of time on arbitrarily large physical
scales. To see this, imagine that the rate is inhomo-
geneous at early times because bubble nucleation starts
at different times in different regions. But since the im-
pact rate depends on the initial time only for bubbles
of radius greater than r
(t, t
) ∼ exp[H(t − t
)], it is
then homogeneous and independent of time on arbitrar-
ily large scales as t → ∞. Third, and for essentially the
same reason, the distribution of inflating region around
any given inflating region also becomes homogeneous and
independent of time on arbitrarily large scales. Fourth,
observers within bubbles lose the information about the
“global time” at which they exist (see Eq. (5)), and all
bubbles become equivalent. Thus the physical descrip-
tion of the universe, relative to any fixed length scale
such as H
, satisfies the Perfect Cosmological Principle
arbitrarily well as t →∞.
B. Eliminating the beginning
The above semi-eternally inflating model can be made
eternal by setting the “state” of the universe to be exactly
that state approached by semi-eternal inflation: because
the statistical properties depend only upon t − t
, for
specified conditions at t
the state at fixed t with t

−∞ is the same as that for t →∞ with fixed t
The state so obtained has the four basic character-
istics listed above: The spatial sections are exactly flat
(outside of the bubbles), the bubble distribution (as char-
acterized by the incoming bubble rate) is homogeneous
and independent of time, as is the distribution of inflat-
ing regions about any inflating region, and the bubbles
are all statistically identical. The inflating region is a
fractal of dimension D < 3 on all scales. This means
that, although inflating regions exist, the global inflating
fraction f
is zero, just as the fraction of 3D Minkowski
spacetime filled by an infinite 2-plane of finite thickness—
an object of fractal dimension two—would vanish. (The
zero probability that a randomly chosen point is in an
inflating region accords with the fact [5] that within the
PLC of each point there is an infinite 4-volume in which
bubbles can nucleate toward that point.) Unlike the re-
gion filled by the plane, however, the inflating region is
statistically homogeneous and isotropic, in that it ex-
actly satisfies the “Conditional Cosmographic Principle”
of Mandelbrot that the statistical description of the in-
flating region about any given inflating region is inde-
pendent of the inflating region chosen (see Ref. [13] for
a discussion of this and other aspects of “cosmological”
This model (essentially derived by Vilenkin [9]) would
seem to have exactly the properties expected of an eter-
nally inflating spacetime, has been straightforwardly con-
structed using the steady-state generated by a semi-
eternally inflating model, and extends to infinite nega-
tive cosmic time. Yet the arguments of [3, 4] (discussed
in more detail in Sec. III) imply that it should be geodesi-
cally incomplete. This issue can be addressed with ref-
erence to the conformal diagram of the model, shown
in Fig. 1. The background inflating spacetime is rep-
resented by the lightly shaded region. Each equal-time
surface is intersected by an infinite number of bubbles
(indicated by light-cones opening toward J
), which are
concentrated along J
and near i
The model thus far constructed (defined in the shaded
region henceforth called “region I”) is geodesically in-
complete: all null geodesics (such as “Y” in Fig. 1), and
all timelike geodesics (such as “X”) other than the co-
moving ones, have only a finite proper time (or affine
parameter) between a point in region I and coordinate
time t →−∞ (see, e.g., [5]). Most geodesics thus “leave
the spacetime” to their past, encountering J

, the limit
surface of the flat equal-time surfaces as t → −∞. (On
the conformal diagram this surface looks like a null cone
emanating from a point at the bottom edge.)
Although the spacetime is geodesically incomplete
there is no curvature blowup or other obvious pathol-
ogy at J

, so the spacetime is extendible rather than
singular. One may take the position that this sort of in-
completeness is allowed, since the edge is outside of the
future of any point in the region, and any given thing in
the spacetime was made at some particular coordinate
time t > −∞
. From this point of view, there is no clear
reason to reject the model as defined in region I.
It seems quite reasonable, however, to ask instead how
the manifold could be extended, and what could be in the
extension. We start by extending the manifold to include

, which is the boundary of the open set comprising
region I. We shall see, as follows, that on J

the field
must everywhere be in the false vacuum. Define φ(λ) as
the field value at affine parameter λ of a non-comoving
geodesic starting at some arbitrary point x
in region I,
where λ increases away from J

. We know that for
some affine parameter λ
, the geodesic encounters J

and also that if our point is within a bubble, there is also
a finite value λ
> λ
at which the geodesic leaves the
bubble and enters the false vacuum φ
. Then for λ < λ
we have
φ(λ) = φ
. (6)
If we then require that the field be continuous along any
geodesic, we then find that the field must be in the false
vacuum φ
everywhere on the surface J

. That is, at
the semi-classical level of description, J

must be an in-
finite null surface of pure false vacuum, through which
no bubbles pass.
Let us now examine the global classical structure of the
background spacetime by momentarily neglecting semi-
classical processes such as bubble nucleations. Then the
manifold comprised of region I and J

is locally dS (con-
stant Ricci scalar R and vanishing Weyl curvature tensor)
everywhere, with φ in the false vacuum (φ = φ
). This
manifold can still be extended past J

, and the obvious
extension is to complete dS spacetime. This is certainly
a solution compatible with our state at J

, and (as we
will argue in Sec. IV and the Appendix) it seems likely to
be unique. Thus we will take the maximal extension of
the background spacetime to be full dS spacetime. That
is, the non-shaded region of Fig. 1, henceforth called “re-
gion II” must simply be the rest of dS spacetime. Con-
sider now a classical field in the background spacetime
obeying a homogeneous hyperbolic equation. Given any
point P in region I, almost all inextendible non-spacelike
curves through P intersect J

. Therefore specifying the
field values on J

effectively poses a “characteristic ini-
tial value problem” [15, 16] with a unique solution ev-
erywhere in region I (this is the analog of the Cauchy
This was the view taken by investigators of the “cyclic
model” [14] which is geodesically incomplete in a very similar
way, and may be the view taken by the adherents of the classic
steady-state model.
FIG. 2: Conformal diagram for eternal double-well inflation.
Bubbles are open FRW regions; equal time slices shown as
curved horizontal lines. For clarity we have not included bub-
ble intersections. Also shown are past light-cones, cut off at

, of both a point P and its antipode −P (note that P
and −P are also reflected across the suppressed two-spheres
in the 4D case).
problem, but with boundary conditions on a null surface;
see Sec. IV and the Appendix for more details). Exactly
the same argument can be made, however, for any point
in region II. Thus specifying classical fields everywhere
on J

determines their values everywhere in dS space-
time. This means that the conditions found to obtain
on J

(by specifying the state in region I and requiring
fields to be continuous) also determine the state in re-
gion II and we can extend our model to region II in an
essentially unique way.
We may now examine the extension of the model to
region II at the semi-classical level by including the bub-
ble nucleations. The form of U(φ) indicates that bubbles
must nucleate
at a fixed rate per unit physical 4-volume.
In region I, this led to an asymptotically steady-state
bubble distribution which, when made exact, implied
that there are no bubbles passing through J

. Thus in
region II, though bubbles must nucleate at the required
rate, none must pass through J

The only way this may occur is, in exact symmetry
with region I, to have a steady-state bubble distribution
on the flat slices of region II, with the bubbles opening
away from J

. This is illustrated in Fig. 2. Region II is,
Bubble nucleation has been perhaps most rigorously analyzed in
Ref. [17], and the boundary conditions for bubble nucleation in
our model correspond exactly to those analyzed in their study.
then, a sort of mirror reflection of region I through J

Thus the answer to the question of how the model can be
extended, and what lies beyond J

, is that an essentially
identical copy of region I lies in region II, connected by
the infinite null surface J

This completes the basic specification of our model.
The cosmology obeys the CP and PCP in that it ad-
mits a coordinatization such that all spatial slices are
statistically homogeneous, and statistically the same as
all others. There is no preferred time in this slicing, nor
is there a cosmological singularity: the model is geodesi-
cally complete. Given the appropriate inflaton potential,
any one of the bubbles could describe our observable sur-
If we consider all bubbles to expand with time, then
Fig. 2 suggests that while in region I the future is to-
ward the top of the diagram (“up”), in region II future
lies toward the bottom (“down”). This leads us to the
issue of the cosmological arrow of time (AOT): why does
the time-asymmetric 2nd law of thermodynamics hold
universally, given that fundamental physics is thought to
admit a symmetry (CPT) that includes time-reversal?
There is some consensus that if this question has an an-
swer, it must ultimately be cosmological, with the time
asymmetry resulting from some qualitative difference be-
tween cosmological “initial” and “final” conditions [18]
that precludes a time-reversal (T-) symmetry of the phys-
ical state in any sub-region of the universe, and hence
induces an AOT.
Although our our model has no “initial” conditions, it
does have boundary conditions on J

(discussed in de-
tail in Sec IV below) and we can discuss the AOT in light
of them. To do so we must divide the universe into two
types of sub-regions: those entirely outside of bubbles,
and those partially or wholely within them. Outside of
the bubbles (or alternatively near enough to J

) there
is no local AOT: the description of such a region admits a
time-reversal symmetry. Were we to hypothesize, for ex-
ample, an imaginary observer outside of a bubble with its
own AOT (pointing away from the time of its creation),
that observer would see only T-invariant dS spacetime.
The observer could not know whether it moved “up” or
“down” on the conformal diagram, nor if it was in region
I or II, nor if it crossed J

. What the observer is guar-
anteed, however, is that it will eventually be encountered
by, and find itself within, a bubble.
The bubble interiors are not time-symmetric: within
a bubble, there is a unique time direction in which the
mean energy density decreases. This direction is away
from the bubble wall/slow-roll inflation epoch, at which
the FRW-region is known to be nearly homogeneous. If
one bubble is to represent our observable surroundings,
this direction must correspond to the time direction in
which the entropy of an isolated system increases. It
FIG. 3: The decay of: (left) an unstable particle, (right) an
unstable vacuum state. In the particle case, the trajectory
is classically describable to good precision at early and late
times, but not near the decay (shaded region). Likewise, the
inflaton φ is classically describable at large invariant distance
∆ from the nucleation event at O, but not near it (shaded re-
gion). This quantum region connects the classically describ-
able field configuration of the bubble interior to that of locally
dS spacetime.
has been often argued, particularly by Penrose [19], that
this connection arises because when gravity is included
homogeneity corresponds to an extremely low-entropy
state. We shall assume this correspondence here (and
that the bubble does not begin in some very special state
for which the density fluctuations decrease). Under this
assumption the physical AOT within any bubble must
point away from the bubble walls; globally this means
that the AOT (where defined) points away from J

As illustrated in Fig. 2, one can therefore indeed draw
timelike geodesics (such as “X”) along which the physical
AOT reverses, but the reversal always occurs to the past
of any physical observer (all of which are within bubbles),
and within a region (the locally dS spacetime) in which
there is no well-defined physical AOT.
We have argued that within bubbles the physical laws,
but not the physical state, admit a symmetry (CPT) in-
cluding time-reversal, while outside of bubbles the laws
and the state admit time symmetry; but what about the
bubble nucleations themselves? Is there not some AOT
telling them “which way” to nucleate, depending upon
which side of J

they are on?
To clarify this point, consider the process of semi-
classical bubble nucleation by analogy to the decay of
an unstable particle, described by a potential V (x) as in
Fig. 3. Classically, one allowed trajectory is for the parti-
cle to sit at constant x in the false minimum x
line on diagram); a second, equal-energy trajectory is
for the particle to “bounce” off of the potential (curved
line in diagram). A purely quantum description of the
system (in the Schr¨odinger representation) would start,
for example, with a Gaussian wave function centered at
at some time t = 0, and hence with the position ex-
pectation value ˆ x ≃ x
. With time, the wave function
spreads, and ˆ x increases, eventually approaching the
classical trajectory for ˆ x ≫ x
(we could say roughly
that the particle has “decayed” when ˆ x changes sig-
nificantly from x
.) Note also that the same spreading
would occur at t < 0 so that this purely quantum de-
scription would be T-symmetric. Now, a semi-classical
description of the system would describe the system clas-
sically at both early and late times, but with a quantum-
mechanical transition connecting the classical trajecto-
ries at some given time. This transition time is random,
and in an ensemble of such systems (as is required for
a correct probabilistic description) would follow a prob-
ability distribution given by a WKB-type calculation of
the decay rate. Near the transition time the system can-
not be described classically; we must “shade out” the
region where only a quantum description is accurate, as
in Fig. 3.
Bubble nucleation can be described semi-classically in
a similar way. Here, we must attach allowed classical so-
lutions of the field equations along some boundary that
represents the nucleation event. To do this in a covari-
ant way, this boundary must be a surface of zero proper
distance, i.e. a null cone, as shown in Fig 3. A bubble nu-
cleation “event” is thus comprised of a region (shown as
the shaded upper quadrant) where a classical bubble in-
terior solution applies, attached to locally dS regions by
a “shaded out” region (within some proper separation
squared of the nucleation point) where only a quantum
description is valid. To produce a semi-classical descrip-
tion of a spacetime in which nucleation events occur, one
must then populate it with these configurations in such a
way that the nucleation sites are randomly situated and
occur at the correct rate per unit four-volume, while the
classical description, which applies far from the nucle-
ation site, is in accord with the (classical) boundary con-
ditions. When the classical boundary conditions are ours,
given on J

, this yields the bubble distribution indi-
cated in Fig. 2. The (semi-classical) boundary conditions
do, then, control the time direction of bubble nucleation,
not by introducing some locally-detectable AOT, but by
controlling the allowed global configuration of bubble nu-
cleation events.
There is one final “region” in which we can check
the AOT: the entire universe. Interestingly, we here
find that while each bubble nucleation event is non-time-
symmetric, by virtue of the symmetry of the cosmologi-
cal boundary conditions, the statistical description of the
universe does admit a sort of T-symmetry. In Sec. V we
will discuss the possibility of making this symmetry ex-
act via an identification on the manifold. This raises the
intriguing possibility of having a well-defined (and con-
sistent among communicating observers) AOT for all ob-
servers even while the physical laws and the global physi-
cal description of the universe both admit a time-reversal
A. The singularity theorems
Having examined the AOT, we may turn to the sin-
gularity theorems, for the AOT proves crucial in their
analysis. Both the older theorems [3] and the newer theo-
rem [4] assert that if certain conditions are satisfied every-
where within a spacetime, then not all past non-spacelike
geodesics have infinite proper time or affine parameter.
This indicates that the spacetime is either extendible or
contains singularities.
The older theorem poses four such conditions, of which
our model satisfies three, as does region I or II by itself.
The fourth condition is that there exists a point P such
that there is a finite difference in volume between the
interior of the PLC of P and that of any point P

the past of P. This is motivated by double-well infla-
tion, and claimed as necessary for semi-eternal inflation
(the argument being that if this condition does not hold,
then any inflating point will find itself in a bubble at
the next instant, with probability one). We argue that
this condition is not quite necessary: what is required is
that there be a finite volume in the region between the
PLCs of P and P

in which bubbles can nucleate toward
P. Thus if, as in the proof of the theorem, the“past of P”
is taken to be the full volume interior to the light-cone
pointing in the time direction away from which a bubble
nucleated at P would expand, then the fourth condition
applies to region I alone (and correctly implies that it is
extendible). But it would not apply to the full spacetime
(which is neither extendible nor singular), because the
relevant part of the light-cone extends only to J

The argument of the newer singularity theorem [4] con-
sists of the definition of a local “Hubble parameter” H
meant to represent the rate of divergence of neighbor-
ing comoving test particles, along with an argument that
any region in which a suitable average H
of H is greater
than zero along all geodesics must be geodesically incom-
We understand Borde at al.’s argument as follows. One
imagines some timelike or null test geodesic with affine
parameter λ in the spacetime in question, then attempts
to construct a timelike vector field u
(λ) along the test
geodesic into its past such that H
(defined via u
) ex-
ceeds zero. It is shown that this can only be achieved
along some finite affine length of the test geodesic, since
the imposed condition rapidly forces u
towards nullness.
Borde et al. [4] then take their result to mean that an
eternally inflating spacetime is past-geodesically incom-
We take the hypothetical satisfaction of their averaged
Hubble parameter condition for all test geodesics as the
implicit definition of what Borde et al. mean by an eter-
nally inflating spacetime. The logic is that u
might be
independent of test geodesic and simply be the velocity
field of some set of comoving worldlines in the inflating
spacetime. So what the theorem actually implies is that
is it impossible to entirely cover a spacetime with such a
set of worldlines in a way that allows all test geodesics
cutting these worldlines to obey the Hubble parameter
For illustration, let us consider dS space. We note that
as pure dS is a maximally symmetric spacetime, it does
not make sense to regard some parts as inflating and oth-
ers not—one must break the symmetry by adding some
other ingredients, and then frame a discussion in terms
of them. Nevertheless, let us investigate if the geodesics
defined by having fixed comoving coordinates themselves
constitute a u
field satisfying H
> 0. It turns out
that H reduces to the usual ˙ a/a in this situation. In the
closed slicing with metric
= −d

+ sin

, (7)
the coordinates cover all of dS, so u
is globally defined.
However, H < 0 for
t ≤ 0 so H
goes negative there.
Now consider using a single flat or open coordinate patch
to define u
. H
> 0 here (at least for the appropri-
ate choice of time orientation). However, because nei-
ther of these coordinate patches covers the spacetime,
neither does u
. Furthermore, one may choose an infi-
nite number of flat coordinatizations of dS, each with a
different null boundary where the u
construction fails.
This makes it clear that the boundary of a given u
cannot be unambiguously used to define an edge to an
inflating region.
We do not believe that the global existence of a suitable
is necessarily the best definition of what is meant by
an eternally inflating spacetime.
In particular, equat-
ing H with physical expansion entails a tacit assumption
that the physical AOT is everywhere in accord with that
defined by u
. The model we have proposed could be
covered by a congruence of geodesics (those comoving in
two flat coordinate patches covering dS) that would yield
H < 0 in some regions. However, we have argued that
these regions may still be regarded as expanding with re-
spect to the physical AOT defined by the cosmological
boundary conditions.
In summary, both singularity theorems postulate con-
ditions for a region to be “inflating”, and find that such
a region cannot be geodesically complete. However, in-
terpreting these theorems as forbidding eternal inflation
seem to us to require an unwarranted assumption about
the global AOT independent of the cosmological bound-
ary conditions.
Indeed the term “eternal inflation” has been used with a vari-
ety of meanings. For example, the recent paper [20] used it to
describe models that are eternally inflating to the future, but
simply geodesically complete to the past ( ˙ a and/or ¨ a may go
negative there). Such histories, with a globally-defined arrow of
time, seem physically unrealistic with a typical mechanism for
exiting inflation, since this would render them unstable to the
formation of thermalized regions in the putatively eternal early
We have seen how extending semi-eternal inflation to
eternal inflation implies particular behavior on the infi-
nite null surface J

. Here we discuss the converse, de-
scribing how the eternal double-well inflation model we
have described can be specified by a particularly simple
set of cosmological boundary conditions that are imposed
on J

A. On cosmological “initial” conditions
The correct specification even of a complete set of
physical laws does not by itself allow prediction of any
physical system’s behavior; these laws must be sup-
plemented by boundary conditions that suffice to fully
characterize the system being modeled. The Big-Bang
(“BB”) model essentially consists of a set of such bound-
ary conditions for our observable cosmic surroundings:
at some early time, our region was a hot, dense, nearly
homogeneous and isotropic mixture of particles and fields
in thermal and chemical equilibrium, in a nearly-flat ex-
panding background geometry with scale-invariant grav-
itational potential perturbations of amplitude 10
the scale of the cosmological horizon.
While these initial conditions yield predictions in ex-
cellent accord with astronomical observations, they are
deemed by many to be too special: they seem to com-
prise an extremely small portion of some “ensemble of
all possible initial conditions”, using some (generally un-
specified) measure (see e.g. [21]). The theory of inflation
is widely accepted as a way to broaden the range of al-
low initial conditions by funneling a relatively wide class
of physical conditions into satisfactory BB initial condi-
tions [22].
This is perhaps reasonable as a stop-gap measure as
long as the true (pre-inflation) initial conditions are un-
known. But it does not solve the initial condition prob-
lem, for not all initial conditions will give rise to infla-
tion [23], nor is it clear that all of those that do give
inflation will yield a viable big-bang model [21]. Thus
one must still assume that suitable conditions emerge
from the initial singularity; whether this assumption cor-
responds to a “special” or “generic” condition seems ill-
defined without a description of the singularity, which
would presumably require quantum gravity. But unless
quantum gravity completely alters the way physical the-
ories are applied, it is unlikely to yield a unique initial
condition; it will still be necessary to supply a quantum
state (or something similar). Even if it is possible to de-
fine the ensemble of all possible states, the question of
how (and why) a state was “chosen” for our universe can
only be a metaphysical one, and it seems that, just as in
choosing physical laws, we can at best posit a state on
some (possibly symmetrical or aesthetic or philosophical)
grounds, derive its consequences, and compare them to
the observed universe.
The same holds for a non-singular cosmological model
such as that proposed in this paper, the basic structure
of which is well-described by semi-classical physics, and
where there can be no hiding of the cosmological bound-
ary conditions in an initial singularity. We instead spec-
ify a particular set, and analyze them in terms of their
simplicity, symmetry, and ability to correctly generate
a big-bang like region that can describe our observable
B. (Semi-)classical boundary conditions for eternal
It is conventional to pose boundary conditions for a
set of classical fields by specifying the fields (and gener-
ally their time derivatives) on a spacelike surface. This
accords with the intuitive idea of specifying the state at
an initial time. An alternative procedure, more relativis-
tic in spirit because it does not assume a particular time
coordinate, is to specify boundary conditions on a null
surface such as the light-cone of a point. (See the Ap-
pendix and Refs. [15, 16, 24] for treatments of the null
initial value problem.)
Our boundary surface J

is a null surface that, when
drawn on the conformal diagram, looks like the light-cone
of a point V

at the bottom (i.e. “at” the conventional
past timelike infinity) and the light-cone of a point V
the top; see Fig.2). As such, one might expect that one
may determine fields (including the spacetime metric)
throughout the space by specifying them on J

. The
conformal mapping, however, hides the fact that J

not a fully closed light-cone: there is always a nonzero
physical volume (of order H
or greater) on any space-
like surface bounded by J

. We must therefore ask if
specifying fields on J

suffices to fix them everywhere, or
if there is information that may come “though the hole”
at V

and V
. In the Appendix, we argue (using the
Green function for a massive scalar field in dS) that the
fields at V
are irrelevant (as long as they are reasonably
well-behaved), as their effect is infinitely diluted.
We can thus pose boundary conditions for our cosmol-
ogy at the classical level as:
1. There exists an infinite connected null hypersurface

of topology R×S
, on which the 4-dimensional
Weyl tensor vanishes and the 4D curvature scalar
is constant.
2. The inflaton field, with a “double well” potential,
is everywhere in the false vacuum on this surface.
3. On this surface, all classical fields are zero or are
in minima of their potentials. This precludes any
radiation propagating through J

This cosmology is classically very dull, as it is just
de Sitter space everywhere with no dynamics. However,
semi-classical bubble nucleations can, without affecting
the fields on J

, create interesting dynamics by forming
bubbles that open everywhere away from J

, and give
rise to eternal inflation as described in Sec. II and shown
in Fig. 2.
C. Quantum mechanical boundary conditions
Although a fully quantum description of semi-classical
phenomena—such as the nucleation of a bubble—is gen-
erally prohibitively complex, we may hope that, be-
cause our set of boundary conditions is so simple at a
semi-classical level, it might be amenable to a simple
quantum formulation, and it would be pleasing to spec-
ify our boundary conditions in an explicitly quantum-
mechanical way. There are two ways one might think of
doing this.
First, one might consider doing usual quantum field
theory on de Sitter space, but with a particular choice
of vacuum corresponding to our “false vacuum” classical
conditions on J

. This might be done by putting the
field in a Gaussian wavepacket around the false vacuum
at (flat slicing) time t, as in [17], then taking a t →−∞
limit. One might then assume that this is sufficient to
define the quantum state over all of de Sitter space. Al-
ternatively, one could set such an initial state at the
t = 0
slice in global (closed) coordinates (7), i.e. on the slice
through the “throat” of the dS hyperboloid. This surface
could then be boosted (infinitely) to become J

. In ei-
ther case, the procedure would be analogous to putting
an unstable particle into a Gaussian in the false vacuum
at t = 0; away from t = 0 in both time directions, the so-
lution of the time-dependent Schr¨odinger equation would
evolve away from the unstable “initial” state.
With our initial data surface being null, one might at-
tempt to define the quantum state directly on it. This
would be conceptually more pleasing, not requiring the
limit through a sequence of spacelike surfaces as above.
Further, a null surface formulation of initial data is rather
elegant [16, 24]. However, unlike for a spacelike initial
value surface, some points on J

are in causal contact.
It is thus less easy to see how to move from a spacelike
to a null boundary value surface in QFT, because fields
do not commute at lightlike separated points. A sim-
ilar effect occurs in so-called “light-cone quantization”
approaches to QFT in Minkowski space [25], where one
uses infinite null sheets (or lightfronts) instead of equal
time slices, and fields again do not in general commute.
The vacuum state for interacting fields is rather easier
to define in the lightfront approach to QFT than in the
usual spacelike approach to QFT, and one might expect
this to hold in a proper null cone approach to QFT. Un-
fortunately, the authors know of no such formulation of
QFT even in Minkowski space.
Thus while we suspect that our cosmological bound-
ary conditions may correspond to a the specification
of a rather simple quantum state, we leave this diffi-
cult problem for future work and here concentrate on
a semi-classical description (though we return to QFT in
Sec. VA).
D. Discussion of the null boundary proposal
The boundary conditions we have proposed are ex-
tremely simple, in line with the view espoused above that
one cannot avoid making a specific choice for cosmolog-
ical boundary conditions, and that it is then reasonable
to make a choice that is as simple and as highly symmet-
ric as possible, rather than hope to choose a “generic”
boundary condition.
1. Relation to other proposals
Our proposal is novel in that it requires no treatment
of an initial singularity, and in that boundary conditions
are placed on a null boundary that does not correspond
to any particular cosmic “time”. It does, however, have
features in common with other proposals for specifying
the state of the universe.
Penrose’s Weyl Curvature Hypothesis (WCH) requires
that initial singularities are constrained to have vanishing
Weyl tensor, corresponding to a low-entropy state. This,
he argues, gives rise to an arrow of time flowing away
from the initial state. Our proposal is very similar, except
of course that we impose the vanishing of the Weyl tensor
on a lightlike slice running across the universe rather than
at an initial singularity.
The Hartle-Hawking No Boundary Proposal (NBP) is
a prescription for the wavefunction of the universe. It
provides Ψ(h
, φ), the amplitude for a given three-metric
and scalar field configuration on a spatial 3D surface.
The construction is designed to suppress irregular con-
figurations relative to more regular ones, and to thus fa-
vor simple and symmetric states. The notion of temporal
evolution does not appear explicitly. However, one can
associate histories with saddle-point approximations to
the wavefunction, and it turns out that such universes
are often smooth when they are young and small. Thus
the NBP may be taken to imply that, at the semi-classical
level, there should exist a spatial surface in a cosmological
spacetime on which the boundary conditions are partic-
ularly simple.
. Our proposal rather uses a null surface
(but see Sec. VI B below). It would be very interesting to
develop a quantum prescription in a similar vein to the
NBP which naturally leads to simplicity on certain null
In the context of models with open-inflationary potentials, the
NBP seems to favor histories in which the scalar field is every-
where in its true vacuum [26]. The NBP may still be relevant
for inflation with the use of an anthropic constraint, or in a
“top-down” approach to calculating quantum probabilities [27].
Starting off in the true vacuum is however not a problem for
recycling models of inflation, as discussed below.
surfaces. For further details on both the WCH and NBP
see [28].
The tunneling approach to quantum cosmology [29] ar-
gues that the semi-classical universe emerges via a quan-
tum tunneling event. In the context of models with
open-inflationary potentials the proposal suggests that
the universe, when first semi-classically describable, is
most likely to be small and regular, with the field away
from the true vacuum; future-eternal inflation can then
ensue. The tunneling is supposed to have occurred out
of a quantum gravitational chaos so severe as to pre-
clude any space-time description. While our proposal
also leads to inflation from semi-classical boundary con-
ditions, it explicitly avoids any such extreme quantum
gravitational regime.
Over the years Sakharov has discussed various cosmo-
logical models involving time-reversal (and CPT) invari-
ance [30, 31, 32, 33]. In these models, the universe is
generally symmetric across a singular FRW bounce at
t = 0, at which universe assumes an especially simple
state; away from this bounce entropy increases in both
directions of time [31]. In [30] he hypothesizes that phe-
nomena at t < 0 are the CPT reflections of the phenom-
ena at t > 0. In [31, 32] he considers the possibility of
an infinite chain of further bounces or cycles away from
t = 0 in both time directions, and also the possibility
that the minimum-entropy surface might be one of max-
imum expansion rather than a singular one [31, 32]. In a
paper primarily about signature change [33] he alludes to
possibility of non-singular time-reversal around a surface
of minimal radius in a false vacuum state. Our proposal
clearly has parallels with these ideas. We, however, con-
centrate on an infinite non-singular null surface, rather
than a singular spacelike one. Moreover, with the con-
cepts of semi-classical bubble nucleation and open infla-
tion, we are able to provide a relatively complete physical
2. The horizon problem
Inflation was originally conceived as a remedy for
troubling issues concerning cosmological initial condi-
tions [34], thus it is useful to compare how such shortcom-
ings of the HBB model [22] are dealt with in our model.
We focus on what is (aside from the singularity problem)
perhaps the most vexing of these HBB difficulties: the
horizon problem.
The horizon problem is generally framed as fol-
lows: choose two spatially antipodal points on the last-
scattering surface. They are similar in temperature, yet
their PLCs never intersect in a HBB cosmology, so there
can be no causal connection between them. Inflation is
generally thought to solve this problem, because with suf-
ficient inflation, the PLCs will intersect. But this does
not suffice for the points to have the same temperature,
because the temperature at each point depends on data
across its full PLC, and there is a portion of each PLC
that does not intersect the other. Taking this into ac-
count, for the two points to have similar properties it
must be assumed that there is sufficient inflation, and
additionally that at inflation’s beginning at time t
, the
region to the past of the two points is homogeneous on
length scales of order H
. But suppose there is a earlier
epoch between some time t and t
. Then at t the patch
must be homogeneous over a physical length scale of at
r(t) ∼
(t), (8)
where H(t) ≡ ˙ a/a. If a ∝ t
(α = 1/2 for radiation-
dominated expansion) prior to t
, then r/(H
) ≃
. Thus it is necessary to postulate that
our region was, at some initial time t
, homogeneous over
Hubble volumes. The horizon prob-
lem therefore persists if t
> t
and α < 1. There are
two escapes available within inflation. The first is to set
= t
= t
, so that the expansion is inflationary all
the way back to the Planck time, before which one can-
not speak of the expansion at all. The horizon problem
is then greatly ameliorated, as it must only be assumed
that some regions “emerging” from the Planck epoch are
homogeneous over a relatively small (but > 1; see [23])
number of Hubble volumes when they are first classically
describable; but whether this constitutes a true solution
will be seen only if and when the Planck epoch itself is
understood. A second potential escape is to set α ≥ 1, or
(more generally) for ˙ a to be non-decreasing, i.e. to have
past-eternal inflation.
How, then, does the horizon problem look in the con-
text of our eternal model with no initial time? Let us pose
the problem in a slightly more general manner: when
specifying the cosmological boundary conditions, must
one do so over a region that is very large compared to
some relevant physical volume such as H
? Posed this
way, it might seem that because J

is an infinite sur-
face, a strong horizon problem exists.
This, however, is
not necessarily the case. Imagine J

as the limit of a se-
quence of spacelike slices obtained by boosting the space-
like surface given by
t = 0 in global coordinates. Because
each such surface has volume 2π
, we might also at-
tribute to J

the same finite invariant volume.
our construction would seem to ameliorate the horizon
This is apparently the case, for example, in the cyclic model [14],
where if the cyclic behavior is to continue indefinitely into the
future and past, it seems necessary to place cosmological bound-
ary confitions on an infinite spacelike surface. This specification
accords the same properties to points that are arbitrarily distant,
and causally disconnected.
One can also consider other ways of taking a limit to J

, but
this one seems most in accord with the symmetries of dS. An-
other approach to the problem is to rather consider some dis-
tance measure between any two points on J

. A dS-invariant
quantity, in units where H = 1, between points P
and P
the embedding space (see Eq.(9) below) is given by D(P
, P
) =
problem (posed in terms of Hubble volumes within the
boundary condition surface) to approximately the same
degree as does inflation beginning in the Planck epoch,
and much better than does inflation with an early quasi-
FRW phase.
The model we have proposed consists of two indistin-
guishable regions, each comprising an eternally inflating
universe with an AOT (where defined) pointing away
from an infinite null surface which connects the two re-
gions. The statistical identity of these universes, along
with lack of a global physical time orientation, suggests
some form of an old idea concerning dS, called “the ellip-
tic interpretation”
that would identify the two universes.
The idea consists of deeming an event to be rep-
resented not by a single point of a spacetime mani-
fold, but by a pair of antipodal points (defined below).
This corresponds to a topological identification that,
applied to our model, identifies regions I and II, and
maps J

onto itself (the R × S
manifold J

(R × S
). This identification has been subject of
some previous [36, 37, 38, 39] and recent [40] investiga-
Pure dS can be represented as a 4D hyperboloid
= H
embedded in 4+1 D Minkowski space with metric
= −dv
. (10)
The elliptic interpretation consists of identifying each
point P with coordinates (v, w, x, y, z) with its antipode
−P at (−v, −w, −x, −y, −z). In the conformal diagram,
this means that points such as P and −P of Fig. 2 are
physically identified; the antipodal map A looks like a
vertical reflection and a horizontal shift through one half
of the diagram’s horizontal extent (Note that although
this map makes an orbifold of the embedding space, there
[11]. In general −∞< D < ∞,
whereas for two points on J

, |D| ≤ 1; they may thus be con-
sidered “close”. The relation between D and geodesic distance is,
unfortunately, not always defined. The points are timelike or null
separated, respectively, for D > 1 or D = 1. For |D| < 1 points
are connectable by a spacelike geodesic of length cos
(D) < π;
but for D ≤ −1 there is no geodesic connecting P
and P
, al-
though one does connect P
to the antipode
of P
) = −D(P
, P
). Under the identification of antipodal
points (see below), any P
and P
are null-separated or spacelike-
separated by a geodesic distance < π/2H.
The etymology of this term is obscure to the authors. The el-
liptic view applied to spatial sections of dS was discussed first
by de Sitter, who preferred it to the (now) conventional “spher-
ical” view; and cited a letter from Einstein voicing the same
preference [35]. Antipodally identifying in space and time was
discussed in fond detail by Schr¨ odinger [36].
FIG. 4: Conformal diagram of dS including Killing vectors
(arrows), a point P and its antipode −P, along with their
light-cones. The causal diamond of an observer following a
geodesic “X” (comoving in the global coordinatization (7)) is
the interior of the dark diamond, the antipodal copy of which
is the light diamond.
are no fixed points in the dS hyperboloid, leaving the
identified space a manifold).
At the classical level, the identification can be enforced
by demanding that all fields in dS are symmetric or anti-
symmetric under A, and that all sources have an accom-
panying antipodal copy. Parikh et al. [40] have argued
that charge conjugation should be added to A, and that
the combination represents CPT. Indeed, in classical field
theories at least, A-symmetry automatically entails op-
posite charges at antipodal points. Consider, for exam-
ple, a complex scalar field, satisfying φ(−P) = φ(P).
Then the global time derivative of the field at the an-
tipode −P is minus that at P, and hence the charge
density (q/2i)


φ − φ∂

) is opposite. How-
ever, in 3+1 D at least, we do not recover the PT part
of their argument. A particle at P with 3-momentum
vector p = (p
, p
, p
) has an antipodal copy at −P. By
an argument like that of Parikh et al., parallel trans-
porting the copy’s trajectory back to P takes the mo-
mentum to (−p
, p
, p
). This looks like parity followed
by a rotation of π about the x-axis. The same proce-
dure, however, takes a small displacement x = (x, y, z)
to (x, −y, −z). The difference in sign arises because the
3-momentum suffers an “automatic” time reversal under
A, whereas the 3-displacement does not. Thus the orbital
angular momentum l = x ×p transforms from (l
, l
, l
to (−l
, l
, l
), just as the 3-momentum does. But under
parity P, x → −x and p → −p, while under T, x → x
and p → −p. Therefore under PT (along with possi-
ble rotations), p and l transform oppositely. Only under
T (with rotations) alone do they transform likewise. So
unfortunately we cannot concur with the beguiling idea
that CPT conjugate events occur at antipodally conju-
gate points in an elliptic universe. Rather, we must settle
for a sort of CT conjugation between processes at antipo-
dal points.
A second interesting (and favorable) feature of the an-
tipodal identification, which holds in pure elliptic de Sit-
ter space (edS), is as follows. An immortal observer in
dS can only eventually “see” (i.e. be connected to via
a non-spacelike geodesic emanating into the past of the
observer) half of the space, the rest being hidden behind
the observer’s dS event horizon. Under the antipodal
identification, however, this “hidden” space is exactly the
same as the “visible” space. Likewise, the space behind
the observer’s particle horizon (the space not reachable
by non-spacelike geodesics emanating from an observer
toward its future) is the same as the space within it. In
this precise sense, edS has no horizons. The notion that
in edS each observer has “full information” about the
space has been a prime motivation for the study of the
space [36, 40].
Along with these appealing attributes however, and
like our cosmological model, edS has some unconven-
tional temporal features that may or may not be desir-
able. Because A includes time reversal, the spacetime is
non-time-orientable: one cannot continuously divide non-
spacelike vectors into two classes which can be labeled
“future” and “past”. Now, one may take the view [36]
that since physics is essentially time-reversible, this poses
no fundamental problem. Non-time-orientability does,
however, have implications for quantum mechanics (see
Sec. VA and [38, 39]). In addition, while physics may
be time symmetric, our physical world manifestly is not,
and this must be confronted in a cosmological model.
The identification of points near “past” infinity with
those near “future” infinity also raises the specter of
closed timelike curves (CTCs) and their accompanying
paradoxes. The identification does not allow any self-
intersecting timelike curves in perfect dS because the full
light-cones of P and −P never intersect (the spacetime
obeys the strong causality principle of [41]). For the same
reason, no observer can see both an event and its antipo-
dal copy. Note, though, that perturbations of dS that
tend to make the conformal diagram “taller” [42] would
allow timelike curves from a point to traverse the dS hy-
perboloid to the point’s antipode.
These temporal features may not be what we are ac-
customed to, but perhaps we should not be surprised
that any cosmology based on globally dS spacetime has
an “unconventional” AOT. It is well known, for example,
that while a patch of dS admits a static coordinatization,
it admits no global timelike Killing vector; as shown in
Fig. 4, the Killing vector is timelike only in half of the
diagram, and moreover points toward the top of the di-
agram only in half of the region in which it is timelike,
reversing completely between antipodal points. (In fact,
the Killing vector field maps onto itself under A).
Our model has much the behavior one would want in
a cosmology based on edS. The physical AOT is defined
only inside the bubbles, which may intersect (and hence
compare time orientations) only within region I or region
II. The two regions are separated by locally dS space-
time, where physics is fully time symmetric. Under the
antipodal identification, the regions are equated, and the
physical AOT is consistent everywhere that it can be
compared by two physical observers. Only an imaginary
observer that could travel “back in time” to leave a bub-
ble, pass through J

, and encounter another bubble,
could see that its own AOT agreed only with that of one
of the two bubbles. (In fact, it would seem that in a non-
time-orientable manifold the physical AOT must either
be undefined in some regions, or must suffer a reversal
along some surface. Thus a construction something like
ours may well be necessary in any cosmology based on
However, an antipodally identified version of our model
does not quite share all of the desirable properties of
edS. The bubbles, with a larger curvature radius than
the embedding space, allow the connection of antipodal
(and hence identified) points by a timelike curve. These
self-intersecting timelike curves (SITCs) are not however
CTCs of the usual grandfather paradox sort. To follow
such a SITC an imaginary observer would, for part of
its journey, have to travel backward in (bubble) time.
Moreover, when the two branches of the SITC meet, they
have opposite time orientation as defined by an affine
parameter along the curve. One might avoid these SITCs
if the bubbles have a smaller curvature radius than the
background space (as discussed below in Sec. VI). In
this case, however, horizons would return, because there
would be regions outside of an observer’s horizon that
are not identified with any region within the horizon.
While the elliptic interpretation is complicated by the
presence of bubbles, the background space of our model is
pure dS, and does benefit from the elliptic interpretation;
all points on J

, for example, would be connectable by
geodesics and have a maximal spacelike geodesic separa-
tion π/2H, and the volume of the boundary condition
surface would be halved.
The elliptic interpretation was suggested by a symme-
try in the statistical description of the bubbles, and it is
interesting to ponder the connection between this statis-
tical symmetry and A. Classically, we may have an en-
semble of systems each of which is not A-symmetric, even
while the statistical properties of the ensemble are. With-
out bubble nucleation, our universe (dS) is A-symmetric.
The boundary conditions which determine the bubble
distribution are A-symmetric, therefore it seems neces-
sarily true that the statistics of the bubble distribution
are A-symmetric. Given quantum mechanics, however,
it is not clear how to relate a single member to the sta-
tistical properties (given by the wavefunction) without
bringing in measurement theory, and we will leave the
question aside for future consideration.
A. QFT in edS
The non-time-orientability of edS also makes quantum
field theory on the space rather more subtle than in usual
dS. We hope to treat this subject in some detail in a
forthcoming paper (see also [38, 39, 40]). Here we sketch
a brief summary.
Consider a massive free scalar field φ(x) obeying some
wave equation, for which we would like to construct a
QFT in edS by defining a Hilbert space of states (in-
cluding a vacuum state |0), and the two-point func-
tion 0|φ(x)φ(y)|0. The latter can be decomposed
into a commutator D(x, y) and an anti-commutator (or
“Hadamard”) function G
(x, y). Under the antipodal
identification, we would expect both D and G
to be
symmetric in some sense under the exchange of x and/or
y with its antipode. How, then, might we define such a
QFT? There are a number of ways, none of which seem
entirely satisfactory.
First, one might just pick a particular “antipodally
symmetric” vacuum state of full dS. Indeed, taking the
α → ∞ limit of the “α-vacua” appropriate for dS [43]
does yield an A-symmetric G
. However, this does not
have the usual short distance behavior of the Minkowski
2-point function. In addition, the commutator D is inde-
pendent of the state chosen, and has no antipodal sym-
metry. (One might also hope to find an A-symmetric
non-vacuum state with the correct short distance behav-
ior, but this would still have the wrong commutator.)
A second approach would be to try to build an an-
tipodally symmetric vacuum in a full dS background, by
choosing a (global) time coordinate and decomposing the
fields into global positive-frequency modes that are an-
tipodally symmetric. The Fock vacuum would then as
usual be the state destroyed by all annihilation opera-
tors. The problem with this approach is that any an-
tipodally symmetric mode turns out to have vanishing
Klein-Gordon norm when integrated over a Cauchy sur-
face for all of dS [38], and the Fock construction breaks
A third approach employed in the literature [38, 39,
40] is to abandon the hope of a Fock vacuum, and to
simply enforce the antipodal symmetry at the level of
the Green functions, by writing antipodally symmetrized
versions of the fields, and computing the resulting two-
point functions in terms of the two-point functions of
unidentified dS. One problem with this approach is that
it becomes somewhat unclear why the anticommutator
should take this value, as the construction seems to lack
an underlying quantum-mechanical motivation. Another
problem is that while the anticommutator so obtained
is antipodally symmetric, the same procedure yields a
commutator that vanishes identically.
A fourth approach is to define the vacuum in terms
of a mode decomposition over only part of dS (such as a
“causal diamond”[11]), where the modes can be consis-
tently positive or negative frequency (so that a Fock rep-
resentation exists), then provide a prescription for defin-
ing correlators between any two points in the space in
terms of correlators within this region. This approach
turns out to be promising; for one or both points in the
causal diamond or its antipodal copy, and for a particular
choice of state (a mixed thermal one at twice the usual
de Sitter Hawking temperature), both D and G
can be
made to have the correct symmetry. The problem arises
when both points are outside of the causal diamond and
its copy; in this case the commutator turns out to van-
ish for timelike separated points, and not for spacelike
separated points. It is unclear whether this makes sense.
In short, defining a satisfactory QFT in edS is rather
difficult; the difficulties stem primarily from the commu-
tator function, because it is not symmetric under time-
reflection, while the anti-commutator is. It is then diffi-
cult for both functions to be symmetric under A, which
includes time-reflection. It is possible [40] that string
theory in edS will make more sense than in dS. It is also
conceivable that the elliptic view could emerge from a
correct quantum treatment of dS, and that the described
troubles stem from doing QFT on a fixed background not
included in the dynamics. This would be an interesting
issue to pursue in string theory or other theories of quan-
tum gravity. For now we must leave it there, and return
to eternal inflation.
We have constructed our eternally inflating universe
using an “open inflation” double-well potential and de-
manding that the inflaton φ rest in the false vacuum
everywhere on the infinite null surface J

. But these
choices are not unique, and the general principles of our
construction can be extended to models employing differ-
ent potentials, or different boundary condition surfaces.
A. Different inflaton behaviors
A simple change in our model can be induced by leav-
ing V (φ) fixed, but demanding that φ rest in the true,
rather than false, vacuum on J

. Although the field is
now in a (positive) stable vacuum, bubbles of false vac-
uum may still be able to nucleate [44]. Assuming this
indeed occurs, the effect of each bubble can be enclosed
within a light-cone, and the distribution of these light-
cones is essentially the same as for bubbles of true vac-
uum, so all of the arguments of Sec II go through. Now,
within these bubbles of false vacuum new bubbles of true
vacuum can form, one of which could describe our ob-
servable universe. At late (bubble) times each bubble
interior would approach dS, dominated by the true vac-
uum (presumably of the magnitude we currently appear
to observe), in which false vacuum bubbles may nucleate
(and so on, ad-infinitum). This is a realization of the
“recycling universe” of [45]. Because bubbles are nested
infinitely deep, the structure of this universe appears ex-
ceedingly complicated. But the global structure at the
outermost level of bubbles is known (and is as in Fig. 2),
because we have specified it using explicit cosmological
boundary conditions. One might argue that this scenario
would have yet simpler boundary conditions than the sce-
nario in which the inflaton is placed in the false vacuum
on J

. Since the lowest vacuum state must be posi-
tive for this scenario to work, it also has the potential
to connect with the presently observed positive vacuum
A somewhat similar scenario can be realized using
“chaotic inflation” potentials such as V (φ) = λφ
+ V
Here we set the inflaton to rest at φ = 0 on J

, and
require V
> 0 so that the space is locally dS near J

This is, again, a stable configuration, yet regions of large
potential may “nucleate” in this dS background as highly
improbable quantum fluctuations in the field. If such a
region nucleates at high enough φ, it would provide the
potential seed for the sort of eternal inflation envisioned
by Ref. [46], in which quantum fluctuations in φ domi-
nate over classical rolling so that inflation become eter-
nal. While the structure of the resulting region becomes
extremely complicated, it can again be contained within
a light-cone of some point in the original (dS) embedding
space, so that the global structure of the universe (filled
with such light-cones) is still understandable, and again
looks like Fig. 2.
B. Different boundaries
We were led to the null surface J

by constructing an
eternal model with a time-translation symmetry. But the
same sort of boundary conditions we apply on J

also be applied to a spacelike initial surface, such as the
t = 0 surface in the global coordinatization (7) of dS.
Such boundary conditions might be closely tied in to the
Hartle-Hawking NBP, as discussed above. The universe
now has a preferred (and initial) time, and constitutes
“semi-eternal” inflation in both the
t < 0 and
t > 0
regions. The same arguments concerning the AOT apply
here: it is undefined near
t = 0 (outside all bubbles),
and defined in bubbles, pointing away from
t = 0. If the
initial surface maps onto itself under the antipodal map,
we can apply the antipodal identification to the universe.
This model bears a stronger resemblance to the earlier
ideas of Sakharov [30, 31, 32, 33] than does our proposal
using the null J

Spacelike surfaces are easily deformable into other
spacelike surfaces, whereas the same is not true for null
surfaces (see e.g. [47]). Thus spacelike surfaces are in
some sense less constrained than null ones and thus
maybe less appropriate in any attempt, such as ours,
to specify cosmological boundary conditions in the most
economical way. In addition, the resulting universe would
not obey the Perfect Cosmological Principle, having a
preferred time at t = 0. As discussed above, we also
might conjecture that the quantum state corresponding
to null boundary conditions is simpler. In general, how-
ever, there seems to be no strong argument against such
a boundary condition surface as compared to a null sur-
We have investigated the possibility of making “future-
eternal” inflation eternal also to the past. Starting with
a de Sitter spacetime background dominated by the false-
vacuum energy in which true-vacuum bubbles can form,
we specify that the bubble distribution at any time is
in exactly the steady-state configuration asymptotically
approached by semi-eternal inflation. All bubbles are
then equivalent, and the statistical distribution of bub-
bles admits both a time-translation and space-translation
invariance in the background inflating space.
The described region (“region I”) has no initial time,
but has a null boundary J

that is the limiting surface
as t → −∞ of the flat spatial sections. The steady-
state configuration at t > −∞ corresponds to the infla-
ton field being in the false vacuum state everywhere on

, with no incoming radiation from J

. This sur-
face can be reached by a past-directed geodesic of finite
proper length from any point in the space, so region I is
geodesically incomplete; this is the “singularity” pointed
to by theorems purporting that past-eternal inflation is
impossible [3, 4]. But the space can (and should) be ex-
tended, as the state on J

also constitutes boundary con-
ditions for the region past J

if the manifold is extended.
These boundary conditions imply that a duplicate copy
of region I exists on the other side of J

. Together re-
gion I and the new “region II” constitute a geodesically
complete (i.e. non-singular) inflating cosmology that ad-
mits a coordinatization in which the background space
and bubble distribution are time-translation and space-
translation invariant.
The null surface J

constitutes a cosmological bound-
ary condition surface on which the universe is classically
in a particularly simple and symmetric state. Although

is an infinite null surface, some of its points are in
causal contact, and one might attribute a finite invariant
volume of O(H
) to it. Therefore specifying boundary
conditions on it is not like specifying them on an infi-
nite spacelike hypersurface (which would lead to a severe
“horizon problem”).
It might be possible to construct a quantum state cor-
responding to our classical boundary conditions on J

taking a null-limit of spacelike sections on which the wave
functional describing the fields is centered on the desired
classical state. But an explicitly null quantum formula-
tion of our null boundary proposal has not been provided
and would constitute an interesting future study.
It is widely thought that the “arrow of time” is con-
nected with cosmological boundary conditions. In our
model, time flows away from J

, and the AOT is con-
sistent among all physical observers that can compare it.
The AOT is not, however, defined globally, and in our
model the statistical description of the universe admits a
global symmetry that includes time-reversal.
This symmetry, along with the presence of two dupli-
cate, non-communicating universes, motivates—though
does not require—formally identifying antipodal points
on the manifold. We, and others, have studied this iden-
tification on de Sitter space classically and quantum-
mechanically. Antipodally identified (or “elliptic”) dS
has the virtues that it is causally stable and observers
have no event horizons. It has the disadvantage that its
non-time-orientability makes defining a reasonable quan-
tum field theory difficult. With antipodal identification
our model is more economical as the two duplicate uni-
verses are identified; however not all of the attractive
features of “pure” edS remain.
Our model can be generalized to other inflaton po-
tentials (such as for chaotic inflation), and therefore al-
lows one to partially understand the global structure of
the eternally inflating spaces that result. One may also
use an analogous construction to specify a semi-eternally
inflating but non-singular universe by placing boundary
conditions on a spacelike section.
Our primary conclusion is that it is possible, using only
“standard” ingredients underlying popular models for in-
flation, to specify simple cosmological boundary condi-
tions on an infinite null surface that lead to past- and
future-eternal inflation. Such a universe would obey the
same Perfect Cosmological Principle that governs a semi-
eternally inflating universe long after its beginning. If our
construction survives scrutiny, and can be specified at the
fully quantum (or quantum-gravitational) level within a
theory of fundamental physics, it could serve as the basis
for a realistic cosmology that avoids a cosmological sin-
gularity, a beginning of time, or a creation of the universe
“from nothing”.
We thank Alex Vilenkin for helpful comments on a
draft of this paper. AA is supported by a grant from the
W.M. Keck foundation. SG is supported in part by US
Department of Energy grant DE-FG02-91ER40671.
Here we review Green functions and the initial value
problem in some detail, focusing for simplicity on the
inhomogeneous scalar wave equation in fixed background
spacetimes. We are interested in how much information
must be specified, and where, in order to fix the field
throughout spacetime.
1. Defining Green functions
Imagine that our scalar field φ satisfies


φ = q (x
) , (11)
where q is an arbitrary source term, and we wish to de-
termine φ at some point P with coordinates x
. This is
possible given φ and its time derivative on some complete
spacelike slice through the past light-cone (PLC) of P.
Let us see why this is so, and whether this is the only
suitable set of initial data for the problem.
FIG. 5: Possible choices of V for different Green functions.
The solid lines indicate the regions S of ∂V which contribute
to the integral. On the left, we have volumes suitable for
the retarded and advanced Green functions GR and GA, and
on the right we have volumes suitable for the commutator
function GC.
First introduce another function G, which depends on
P and is assumed to satisfy


G = s (x
) , (12)
where s is to be chosen. Taking G times (11) minus φ
times (12) and integrating through some four dimensional
volume V yields




φ −φ




−g (qG−sφ) d
x. (13)
This allows determination of φ(P) in terms of its values
elsewhere by a suitable choice of s, G, and V.
To effect this, we must isolate φ(P), either on the left
or right hand side of (13). Let us first use the RHS. Take
s to be a δ-function at P (or, more carefully, a function
peaked near P that can be taken to a δ-function limit).
Then choosing V to enclose P gives φ(P) for the second
term on the RHS of Eq. (13). Using Gauss’s law, this
equation can now be used to express φ(P) as a surface
integral over of φ and its derivative over the boundary ∂V
of V, plus a volume integral over V of the source alone.
To fix G(x), we must choose boundary conditions spe-
cific to our choice of s. Two conventional choices are to
require G to vanish either to the future or to the past of
P. In the first case G is known as the “retarded” Green
function and may be denoted G
; in the second case it
is “advanced” (G
We must now choose V (see Fig. 5). It turns out (see,
e.g., [15]) that G
= 0 only on and within the FLC of
P, and that G
= 0 only on and within the PLC of P.
Hence only a segment S of ∂V contributes to Eq. (13),
enabling us to deduce φ(P) using only data on some con-
nected surface making a complete span through either the
FLC or the PLC of P. The usual choice, consistent with
our standard ideas about causality, is to pick a spacelike
surface of constant time (see left side of Fig. 5). Then
evaluation of Eq. (13) requires φ and
φ on S. For a more
general ∂V , we require φ and its normal derivative on S.
(Note that this surface need not be everywhere spacelike,
but if not then the data must be self-consistent.) The
case of present interest is that of a null surface, for which
the normal lies everywhere within the surface itself (this
is made possible by the Lorentzian signature of the space-
time.) Then specifying φ on the surface also specifies the
normal derivative of φ on it. A suitable null surface is,
for example, the forward light-cone L of a point Q to the
past of P; now we need only know φ on the surface S
where this FLC of Q is within the PLC of P (see right
side of Fig. 5). (If the data near Q is suitably regular, the
non-smooth nature of the surface at Q is unimportant.)
In the same spirit, one might consider a “wedge” such as
the boundary of the future of the segment of a line in the
past of P.
Let us now consider the second way of isolating φ(P),
using the LHS of Eq. (13). Set s = 0, and choose some
piece of ∂V to be spacelike and run through P. Then
choose G to vanish everywhere on this surface. Further,
choose it to have a δ-function at P in its normal deriva-
tive; the integral then picks up a contribution propor-
tional to φ(P). This choice comprises boundary condi-
tions for G, and the function they determine, which we
denote G
, now vanishes outside the (full) light-cone of P
and satisfies the homogeneous version of Eq. (12). Choos-
ing now the remainder of ∂V to close off to the future or
to the past of P, we again may determine φ(P) once data
is given on a complete slice S through one of the light-
cones. G
may be called the commutator function, be-
cause it turns out to be equal to −i times the commutator
of a free quantum field (at least in a globally hyperbolic
spacetime). Note also that G
= G
− G
: subtract-
ing the (inhomogeneous) equations governing G
and G
leaves a homogeneous equation for the difference. Then
the difference must vanish outside the light-cone of P,
just as for G
In all cases, G depends on the point P, and so may
be considered as a function of two variables, x and x
Because φ satisfies (11), it turns out that the G also
satisfies (11) with respect to x
, at least away from x
2. Green functions for Minkowski and dS
Let us now outline a procedure to obtain some Green
functions for massive scalar fields in certain spacetimes,
using a slightly unconventional but perhaps more intu-
itive method not relying on the usual Fourier techniques
or iǫ-prescriptions. We start with the commutator func-
tion for a massless field in 3+1 D Minkowski space,
= sgn(t) δ


/2π, (14)
with coordinates such that P is at the origin. That
this solves φ = 0 can be seen by expanding out the
δ-function and comparing to a general superposition of
incoming and outgoing spherically symmetric waves. (A
nice discussion of this function is found in [48].) Note
that this is only non-zero on the light-cone itself, vanish-
ing even inside the cone. This is a special property of
massless fields in even-dimensional spacetimes (see [15]
for more details.)
To obtain the solution for the massive field in
Minkowski space, we start by considering
≡ f


sgn (t) Θ


The sgn and Θ functions enforce the gross features of
the behaviour that we require. Writing using spherical
polar coordinates, we find
( −m
= sgn(t) Θ



− 4f (0) sgn(t)δ


. (15)
Let us choose f(τ
) (where here τ
= t
− r
) such that


f = 0 inside the light-cones; a general solution
is f(τ
) = AJ
(mτ)/τ +BY
(mτ)/τ where J
and Y
Bessel functions as in Ref. [49]. The first term of Eq. (15)
is then zero. For the second term, we notice that it is
basically the solution to the massless problem that we’ve
already found. So let us simply add the massless solution
from Eqn. (14) to our ansatz G
. Since G
= 0,
we have left −m
on the RHS. By choosing A =
−m/4π and B = 0, we obtain a complete cancellation,
leaving us with our desired solution. Thus the full Green
function is
sgn (t)




. (16)
We may now consider generalizing Eq. (16) to other
spacetimes such as dS. First, let us note that we took
f above to be a function of the proper time τ from the
origin, within the light-cone. Thus the commutator func-
tion was invariant under Lorentz boosts. This suggests
that in dS, for example, we make our commutator inside
the forward light-cone invariant under isometries which
leave P fixed. This is the same as saying that it should be
a function of the proper time from the origin alone. One
can define this quantity in terms of an “angle” in the 5D
embedding space [11] (see footnote to Sec IVD2), but
perhaps a more intuitive way to proceed is as follows. A
patch of dS can be covered by coordinates in which the
metric is the same as that for an open expanding FRW
universe, with scale factor H
sinh Ht. The proper time
along the geodesics representing comoving observers is
just given by the coordinate time t. These geodesics all
intersect at a point as t → 0, which we choose to be
P, and they cover all of the interior of the FLC of P.
We therefore need only find an appropriate spatially ho-
mogeneous solution of the massive wave equation in the
coordinate patch of the open slicing of dS. The equation



φ = 0. (17)
To solve this [50], write φ = Hχ/ sinhHt and set z =
coshHt to obtain
(1 −z


2 −

1 −z

χ = 0, (18)
which is Legendre’s equation [49] with ν = −1/2 +

9/4 −m
and µ = −1. The solution to this equa-
tion which is regular as z → 1 is P
(z). In terms of t,
this tends to Ht/2 as t →0 independent of m
, hence φ
tends to H/2.
We can now use this result to deduce the form of the
commutator over all of dS space. Extend the meaning of
to be the signed geodesic distance squared between P
and the point in question. Note that we do not consider
the point to be in the light-cone of the antipode of P,
since no geodesic exists which connects it to P. In this
case we rather define G
to be zero (but see our discus-
sion of the antipodal quantum commutator above.) Near
P the space is locally Minkowskian, so we can compare
to our Minkowksi result to obtain:

) −
2 sinhHτ

Here by t is meant some suitable generalization of
Minkowski time, with sgn(t) thus serving to make G
be of opposite sign in the future and past light-cones, and
τ = +

. The sgn(t)δ(τ
) should be interpreted as the
generalization of the equivalent term in the Minkowski
3. Domains of dependence
Given our Green functions, we would like to investigate
to what extent fixing the fields on a null-cone L of a
point Q determines the field values within that cone. We
are particularly interested in the importance of the field
near Q, as our cosmological boundary surface J

be considered the light-cone with Q “at” past-timelike
Take, as a first example, the Green function G
for a
massless field such as the scalar field in 4D Minkowski
space with G
from Eq. (14). Because G
has support
only on the PLC of P, the field at P depends only on the
intersection of P’s PLC and L. This indicates two things.
First, specifying data on L explicitly determines the field
everywhere inside L (i.e. everywhere in the future of Q).
Second, the field at any point P inside L does not de-
pend on φ(Q); we might thus consider Q to be irrelevant
in terms of what occurs within L. This, however, hides
a subtlety: while G
and φ on L always allow the con-
struction of a valid solution of the field equations, there
is no guarantee that the field so obtained is continuous
with the field specified on L unless some assumptions of
field regularity at the vertex (Q) are made: if we wish
to evaluate φ on L to check that we have a genuine solu-
tion to our boundary value problem, then we must either
know φ(Q), or assume that the field is regular as Q is
approached along L, so that we can extend the field to
Q by continuity.
For a massive field the strict “Huygen’s Principle” does
not hold; while much of the contribution to the integral
in Eq.(13) comes from the PLC of P where G
is singular
(see [15] for some discussion of this “generalized Huygen’s
principle”), there is also a contribution from inside P’s
PLC because G
is everywhere nonzero there.
In the dS case, G
as given by Eq. (19) falls off expo-
nentially as τ →∞; thus we may expect that when L is
taken to be J

, where the vertex lies “at” τ = ∞, the
field values inside J

will not depend on the field at the
vertex (where by “at the vertex” we mean within J

the limit τ → ∞). More explicitly, we may consider a
boundary surface comprised of J

t >
for some
, closed by the spacelike surface
t =
, where
t and
are in the global closed coordinates. Then the field
integral (13) consists of an integral along J

, and an
integral over the
t =
hypersurface, which has a finite
physical volume of order H
. But then as
→ −∞,
τ → ∞, so this second contribution will vanish unless
either the average of the field or its
t-derivative blow up
faster than G
. Hence we expect that the field at any
point in the space will depend only on the values on J

and not on the field behavior at past infinity, as long as
the fields are assumed to be suitably finite and regular.
This argument might also be applicable to gravity; the
equations governing the Weyl conformal tensor can be
cast, at the linear level, as wave-equations for a spin-2
field, and boundary conditions can be specified on a null
surface such as J

[24]. We therefore expect that our
boundary conditions on J

determine the Weyl tensor
uniquely, and therefore determine the spacetime to be
pure dS when bubble nucleations are neglected.
[1] A. Linde, D. Linde and A. Mezhlumian, Phys. Rev. D
54, 2504 (1996); A. Vilenkin, Phys. Rev. D 46, 2355
[2] See, for example, A.H. Guth, Phys. Rept. 333, 555 (2000)
and references therein; A.D. Linde, Phys. Lett. B175,
395 (1986); A.D. Linde in The Very Early Universe,
edited by G.W. Gibbons, S.W. Hawking and S.T.C. Sik-
los (Cambridge University Press, Cambridge, England,
1983); P.J. Steinhardt, same volume.
[3] For a review, see A. Borde and A. Vilenkin, Int. J. Mod.
Phys. D5 (1996) 813-824.
[4] A. Borde, A.H. Guth and A. Vilenkin, gr-qc/0110012.
[5] A. Aguirre and S. Gratton, Phys. Rev. D 65 (2002)
[6] A.H. Guth, Phys. Rev. D 23, 347 (1981).
[7] M. Bucher, A.S. Goldhaber and N. Turok, Phys. Rev. D
52, 3314 (1995).
[8] J.R. Gott, Nature (London) 295, 304 (1982); K. Ya-
mamoto, M. Sasaki, and T. Tanaka, Astrophys. J. 455,
412 (1995); A. Linde and A. Mezhlumian, Phys. Rev. D
52, 6789 (1995)
[9] A. Vilenkin, Phys. Rev. D 46, 2355 (1992).
[10] A.H. Guth and E.J. Weinberg, Nucl. Phys. B212, 321
[11] For reviews of de Sitter space, see: Ref. [41]; M. Spradlin,
A. Strominger, and A. Volovich, hep-th/0110007.
[12] S. Coleman and F. De Luccia, Phys. Rev. D 21, 3305
[13] B. Mandelbrot, The Fractal Geometry of Nature, (W.H.
Freeman and Co., New York, 1983).
[14] P.J. Steinhardt and N. Turok, hep-th/0111030.
[15] R. Courant, Methods of Mathematical Physics: Vol. II,
Partial Differential Equations (New York: Interscience
Publishers, John Wiley & Sons, 1962).
[16] R. Penrose, Gen. Rel. and Grav. 12, 225 (1980).
[17] V.A.Rubakov and S.M.Sibiryakov Theor. Math. Phys.
120, 1194 (1999).
[18] See, e.g., H.D. Zeh, “The physical basis of the direction
of time”, (Berlin: Springer-Verlag, 1999).
[19] R. Penrose, in “Quantum Gravity 2: A Second Oxford
Symposium”, ed. C.J. Isham, et al. (Oxford: Clarenden
Press), 1981.
[20] G.F.R. Ellis and R. Maartens, qr-qc/0211082.
[21] R. Penrose, in “Fourteenth Texas Symposium on Rel-
ativistic Astrophysics, Dallas, Tex., 1988” (New York:
Annals of the New York Academy of Sciences, 1989); S.
Hollands and R. Wald; gr-qc/0205058; N. Turok, Class.
Quantum Grav. 19, 3449 (2002).
[22] A. Linde, “Particle physics and inflationary cosmology”,
(Switzerland: Harwood Acedemic Publishers, 1990).
[23] T. Vachaspati and M. Trodden, Phys. Rev. D 61, 023502
[24] R. Penrose and W. Rindler, “Spinors and Space-Time,
Vol. I”, (Cambridge: Cambridge University Press, 1984).
[25] T. Heinzl, Lect. Notes Phys. 572, 55 (2001),
[26] S. Gratton and N. Turok, Phys. Rev. D 63, 123514
[27] S.W. Hawking and T. Hertog, Phys. Rev. D 66, 123509
[28] S.W. Hawking and R. Penrose, The Nature of Space and
Time (Princeton University Press, Princeton, New Jer-
sey, USA, 1996).
[29] A. Vilenkin, Phys. Rev. D 27, 2848 (1983).
[30] A. D. Sakharov, JETP Lett. 5, 24 (1967).
[31] A. D. Sakharov, Sov. Phys. JETP 52, 349 (1980).
[32] A. D. Sakharov, Sov. Phys. JETP 56, 705 (1982).
[33] A. D. Sakharov, Sov. Phys. JETP 60, 214 (1984).
[34] A. H. Guth, Phys. Rev. D 23, 347 (1981).
[35] W. de Sitter, M.N.R.A.S 78,3 (1917).
[36] E. Schr¨odinger, Expanding Universes (Cambridge Uni-
versity Press, Cambridge, England, 1956).
[37] H. Schmidt, Fortschr. Phys. 41, 179 (1993).
[38] G. Gibbons, Nuc. Phys. B271, 497 (1986).
[39] N. Sanchez, Nuc. Phys. B294, 1111 (1987); J.L. Fried-
man and A. Higuchi, Phys. Rev. D 52, 5687 (1995).
[40] M. Parikh, I. Savonije and E. Verlinde, hep-th/0209120.
[41] S.W. Hawking and G.F.R. Ellis, The large scale structure
of space-time (Cambridge University Press, Cambridge,
England, 1973).
[42] F. Leblond, D. Marolf and R. C. Myers, JHEP 0206, 052
[43] E. Mottola, Phys. Rev. D 31, 754 (1985); B. Allen, Phys.
Rev. D 32, 3136 (1985).
[44] K. Lee and E.J. Weinberg, Phys. Rev. D 36, 1088 (1987).
[45] J. Garriga and A. Vilenkin 1998, Phys. Rev. D 57, 2230
[46] A. Linde, D. Linde and A. Mezhlumian, Phys. Rev. D
49, 1783 (1994).
[47] R. Bousso, Rev. Mod. Phys. 74, 825 (2002).
[48] P.A.M. Dirac, Lectures on Quantum Field Theory, (New
York: Belfer Graduate School of Science), 1996.
[49] M. Abramowitz and I. Stegun, “Handbook of Mathemat-
ical Functions” (New York: Dover Publications, 1972).
[50] M. Sasaki, T. Tanaka and K. Yamamoto, Phys. Rev. D
51, 2979 (1995).

Master your semester with Scribd & The New York Times

Special offer for students: Only $4.99/month.

Master your semester with Scribd & The New York Times

Cancel anytime.